alanhuangyoo d87a1c1193

fix(client): emit text a later node appends to an AI message already sent (#5479) ...

* fix(client): emit text a later node appends to an AI message already sent

Loop detection, the token budget, safety termination, subagent limits and the
terminal-response fallback rewrite the last AI message under the same id in
their own after_model node. stream() skipped every id it had seen, so chat(),
the TUI and --print kept the text from before the rewrite: no [FORCED STOP]
notice, no fallback error, and token_usage_attribution added after the model
node never arrived.

Look again at a known id when a snapshot holds a different message object,
emit only the text appended to what was already sent, and send new
additional_kwargs as the existing metadata-only follow-up.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

* test(client): pin that a non-extending replacement is not re-sent

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5 <noreply@anthropic.com>

2026-09-17 08:45:29 +08:00

..

extension-api

feat(extensions): expose incremental run evidence reader (#5405)

2026-09-16 21:18:05 +08:00

harness

fix(client): emit text a later node appends to an AI message already sent (#5479)

2026-09-17 08:45:29 +08:00